Considerable effort has been made in the last two decades regarding the design and implementation of tools aimed at fast characterization of earthquake sources. As a fruitful harvest, a preliminary magnitude estimation is often available within fifteen minutes for a Mw=6.0 earthquake occurring anywhere in the world. But until recently, it still took several hours to determine the first order attributes of a great earthquake (e.g. Mw≥8.0) even in a well instrumented region. Our goal here is to develop a method providing fast centroid moment tensor (CMT) estimates for large earthquakes based on the W phase.The W phase is a very long period phase (100-1000 s) starting at the same time as the P-wave. It is conspicuous on broadband displacement records before the surface waves arrivals and can then be used for fast characterization of the source. We show that the W phase is not affected by shallow heterogeneities since a significant part of its energy is propagating deep into the mantle. Because of its long period nature, the W phase algorithm is particularly well suited for robust determination of CMT parameters for Mw≥7.5 earthquakes. For large earthquakes indeed, the source complexity is more evident which translates into significant variations of narrow-band magnitude estimations performed at short period. Tsunami earthquakes and outer-rise events provide good examples of this problem (Kanamori, 1972 ; Lay et al., 2009).In the last three years, several international collaborations have been established to use and test the algorithm online either at global or at regional scale. At teleseismic distances, the W phase method allows CMT estimates within half an hour after the earthquake origin time. At the regional scale, the CMT solutions are available between 6 min and 12 min after the origin time. The results obtained in real-time at the United States Geological Survey (USGS), the Pacific Tsunami Warning Center (PTWC), the Institut de Physique du Globe de Strasbourg (IPGS) and in the region of Mexico clearly indicate the robustness and the accuracy of CMT solutions computed using the W phase.Using a bayesian formulation, we propose a formal error analysis for seismic source inversion based on long period seismological data. Taking into account more realistic data uncertainties allows us to improve error estimates on the source model parameters but also to improve the solution itself. In this study, we also develop a method to estimate the uncertainty on the centroid depth. This parameter has a strong influence on the scalar seismic moment and on the fault dip estimated during CMT inversions.
